This invention relates generally to systems and methods for controlling a process and is particularly directed to the analysis of alarms in a process based upon the functional relationships that a newly activated alarm has with other currently activated alarms.
There are generally two types of methodologies in the handling of alarms in process control. One approach makes use of a predetermined, static ranking of the alarms. Thus, all of the alarms in the process are considered as a group and ranked according to their relative importance. These rankings are incorporated within an alarm system such that, no matter what the state of the process, an alarm will always be emphasized according to this predetermined set of rankings. While this methodology can be helpful, it is incapable of adjusting the importance of alarms based on the dynamics of the process being monitored. For example, the importance of any given alarm may be greater under one specific set of conditions and it would therefore be highly desirable to know when this alarm is activated under this specific set of conditions. Under another set of conditions the same alarm may be expected and might actually be the consequence of another alarm or of a process state. In these cases, the alarm should not be displayed at a very high level of importance since it is more of a status indicator than alarm condition.
Another methodology, which has not been widely accepted in industry, utilizes time-ordered sequences to determine the importance of alarms and to perform other diagnostic functions. In this approach, all possible (or likely) alarm activation sequences are identified and modeled. As a given scenario in the process being monitored develops, the alarm sequence is matched to the modeled sequences in attempting to identify what the current and future state of the process is or is likely to be. This approach is generally presented in the form of logic, or cause-consequence, trees. Unfortunately, these logic trees are difficult and expensive to develop and build, are generally inflexible to change, and are not easily maintained over the life of a plant. As a result, the logic tree approach to alarm analysis has been of limited use in real applications. Examples of the logic tree approach, particularly as applied to the environment of a nuclear power reactor, are: DMA (Diagnosis of Multiple Alarms) , disclosed in an article by M. M. Danchak, entitled "Alarms Within Advanced Display Streams: Alternatives and Performance Measures", published in NUREG/CR-2276, EGG-2202, September 1982; STAR, disclosed in an article by L. Felkel, entitled "The STAR Concept, Systems to Assist the Operator During Abnormal Events," published in Atomkernenergie, Kertechnik, Vol. 45, No. 4, 1984, pp. 252-262; and DASS (Disturbance Analysis and Surveillance Systems), disclosed in an article by A. B. Long, R. M. Kanazava et al, entitled "Summary and Evaluation of Scoping and Feasibility Studies for Disturbance Analysis and Surveillance Systems (DASS)", published in Topical Report EPRI NP-1684, December 1980.
The present invention is intended to overcome the aforementioned limitations of the prior art by providing an alarm filtering or analysis methodology based upon the functional relationships of alarms which is not only sensitive to the dynamic nature of the process being monitored, but also is capable of changing alarm importances as necessary. The present invention utilizes artificial intelligence techniques and knowledge-based heuristics to analyze alarm data from process instrumentation and respond to that data according to knowledge encapsulated in objects and rules.